Term 2 Causality revisted Flashcards
What are the ways that you can solve average treatment effect bias?
-RANDOM SELECTION
-IV ESTIMATION
What is propensity score matching?
How can you make the control group look as much like the treated group as possible.
Matches people in treated group, with comparators in untreated group based on observables.
What is regression discontinuity design?
What does it do?
-Looking at those marginally either side of the cut off point
-Runs OLS based on those marginally close to the cut off
-Purely random behaviour near cut off point
What is the average treatment effect of the treated?
Why is the observed difference different to the average treatment effect of the treated
- Because of the selection bias.
How can you work out the direction of the selection bias?
-Try and work out what is higher in the subtraction
What is random selection to avoid selection bias?
What does this cause?
Random selection mechanism like dice that randomly allocates people going to hispital
THe observed difference = average treatment effect.
Then estimate by OLS
How does IV solve selection bias?
Find suitable intrument for Di , which is zi
zi must solve two conditions:
-Relevance Cov(zi,di) is not equal to 0
-Exogeneity zi is not impacted by unobserved error terms.
biv = Y1bar - Y2bar / D1bar - D0bar
Y1 bar = Ybar | Z=1
Y0 bar = Ybar | Z=0
D1 bar = Dbar | Z=1
D0 bar = Dbar | Z=0
Then look at only the compliers and look at the difference and this is the local average treatment effect.
What are the different categories under the IV to solve selection bias?
Compliers, they do what the instrument
Always takers, they always do what the instrument says
Never takers, they always do the opposite of what the instrument says
Defiers, they always do the opposite of the instrument
When filling in a compliers and non-compliers table what is something you to to complete it?
-You cross out defiers
What is the way to interpret the coefficient on an interactive dummy?
You need to do every case then subtrct each one
Then subtract the differences
label each one
How can you graph difference in difference?
Draw a before and after line and draw the lines
What is the assumption for panel data difference in difference
if the before hand data points show that they look the same the next one can be done under the same assumption
-Pattern of something is sufficiently similar
How do you do, difference in difference in difference?
Do the 8 different scenario changing things and subtract them
Number all the different effects
DDD has 4 terms
